/* Tom Gall (tom.gall@linaro.org / tom_gall@mac.com)
 * Copyright Linaro 2014
 * Released under the terms and conditions of the license documented in the LICENSE file
 */
#include <stdio.h>
#include <CL/cl.h>

#include "sqlite3.h"
//#include "sqlite3ext.h"
#include "opencl-sqlite.h"

#define SEARCH_KERNEL_FUNC "x1_search_kernel"

int opencl_init( opencl_sqlite_context *s, sqlite3 *db, 
	size_t data_size,  size_t results_size, char *filename) {
	cl_platform_id platform;
	cl_device_id device;
	cl_int i, err;
	FILE *program_handle;
	char *program_buffer, *program_log;
	size_t program_size, log_size, read_return;

	s->db = db;
	s->data_cpu = malloc(sizeof (opencl_data));
	s->data_size = data_size;
	s->data_cpu->d = malloc (data_size);
	s->data_gpu = malloc(sizeof (opencl_data_gpu));
	s->data_gpu->d = NULL;
	s->results_cpu = malloc(sizeof (opencl_results));
	s->results_size = results_size;
	s->results_cpu->r = NULL;
	s->results_gpu = malloc(sizeof (opencl_gpu_results));

/* Identify a platform */
   err = clGetPlatformIDs(1, &platform, NULL);
   if(err < 0) {
      perror("ERROR: Couldn't find any platforms.");
      exit(1);
   }

   /* Access a device */
#ifdef USE_CPU
   err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_CPU, 1, &device, NULL);
#else
   err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL);
#endif
   if(err < 0) {
      perror("ERROR: Couldn't find any devices.");
      exit(1);
   }

   /* Create the context */
	s->context = clCreateContext(NULL, 1, &device, NULL, NULL, &err);
	if(err < 0) {
		perror("ERROR: Couldn't create a context.");
		exit(1);
	}	

	err=clRetainContext(s->context);	
	if(err < 0) {
		perror("ERROR: Couldn't retain context.");
		exit(1);
	}	

	/* Read program file and place content into buffer 
    * loop here across useful cl routines that we'll have
    */
	program_handle = fopen(filename, "r");
	if(program_handle == NULL) {
		perror("Couldn't find cl file \n");
		exit(1);
	}
	fseek(program_handle, 0, SEEK_END);
	program_size = ftell(program_handle);
	rewind(program_handle);
	program_buffer = (char*)malloc(program_size + 1);
	program_buffer[program_size] = '\0';
	read_return = fread(program_buffer, sizeof(char), program_size, program_handle);
	fclose(program_handle);

   /* Create program from file */
   s->program = clCreateProgramWithSource(s->context, 1,
      (const char**)&program_buffer, &program_size, &err);
   if(err < 0) {
      perror("Couldn't create the program");
      exit(1);
   }
   free(program_buffer);

   /* Build program */
   err = clBuildProgram(s->program, 0, NULL, NULL, NULL, NULL);
   if(err < 0) {

      /* Find size of log and print to std output */
      clGetProgramBuildInfo(s->program, device, CL_PROGRAM_BUILD_LOG,
            0, NULL, &log_size);
      program_log = (char*) malloc(log_size + 1);
      program_log[log_size] = '\0';
      clGetProgramBuildInfo(s->program, device, CL_PROGRAM_BUILD_LOG,
            log_size + 1, program_log, NULL);
      printf("%s\n", program_log);
      free(program_log);
      exit(1);
   }

   /* Create kernel for the all the added cl functions 
    * TODO : evolve this in a loop
    */
   s->kernel = clCreateKernel(s->program, SEARCH_KERNEL_FUNC, &err);
   if(err < 0) {
      perror("Couldn't create the kernel ");
      exit(1);
   }

	/* Create a CL command queue for the device*/
   s->queue = clCreateCommandQueue(s->context, device, CL_QUEUE_PROFILING_ENABLE, &err);
   if(err < 0) {
      perror("Couldn't create the command queue");
      exit(1);
   }

	return OPENCL_SUCCESS;
}

int
opencl_release(opencl_sqlite_context *s) {

	return OPENCL_SUCCESS;
}

int
opencl_cleanup(opencl_sqlite_context *s) {
	clReleaseKernel(s->kernel);
	clReleaseCommandQueue(s->queue);
	clReleaseProgram(s->program);
	clReleaseContext(s->context);

	return OPENCL_SUCCESS;
}

/* 
 * The point of this code is to get at the data in the database so we can work with it.
 * Likely to be replaced at some point by going completely to shards by column which
 * is the most efficient when passing into an OpenCL kernel.
 */
int
opencl_prepare_data(opencl_sqlite_context *s, const char* sql_stmt, int * columnSettings) {
	sqlite3_stmt *stmt;
    sqlite3_value *val;
	int columns, offset=0, type, stride=0;
    opencl_data *data = s->data_cpu;

    int i = strlen(sql_stmt);

    sqlite3_prepare_v2(s->db, sql_stmt, i, &stmt, NULL);

	// first determine the size of a row 
    i = sqlite3_step(stmt);

	// no row? bail 
    if(i != SQLITE_ROW)
        return OPENCL_ERR_STMT;

    columns = sqlite3_column_count(stmt);
    data->columns = columns;

	// walk one row
    for(i = 0; i < columns; i++) {
        val = sqlite3_column_value(stmt, i); 

        // Handling just the 32 bit cases for now
		// will support armv8, other 64 bit hardware eventually
		type = sqlite3_value_type(val);
		switch (type) {
			case SQLITE_INTEGER:
				data->types[i] = OPENCL_INT;
				data->offsets[i] = offset;
				offset += sizeof(int);
				break;
			case SQLITE_FLOAT:
				data->types[i] = OPENCL_FLOAT;
				data->offsets[i] = offset;
				offset += sizeof(float);
				break;
			default:
				perror("Unsupported data type detected\n");
				return OPENCL_ERR_TYPE;
		}
    }

    // set row stride to the current offset
	for ( i = 0; i < columns; i++ ) {
		if (columnSettings[i])
			stride+=4;
	}

    s->data_cpu->stride = stride;
    int rows = 0;
    char *d = (char*)s->data_cpu->d;
    void *p;
	int rowoffset=0;

    do {
        for(i = 0; i < columns; i++) {
			if (columnSettings[i]) {
            	p = d + rows * stride + rowoffset;

            	switch(data->types[i]) {
                	case OPENCL_INT :
                    	((int *)p)[0] = sqlite3_column_int(stmt, i);
                    	break;
                	case OPENCL_INT64 :
                    	((long long *)p)[0] = sqlite3_column_int64(stmt, i);
                    	break;
                	case OPENCL_FLOAT :
                    	((float *)p)[0] = (float)sqlite3_column_double(stmt, i);
                    	break;
                	case OPENCL_DOUBLE :
                    	((double *)p)[0] = sqlite3_column_double(stmt, i);
                    	break;
                	default:
                    	fprintf(stderr, "column %i, datatype is not supported (yet)\n", i);
                    	return OPENCL_ERR_TYPE;
				}
			rowoffset+=4;
            }
        }

        rows++;
		rowoffset = 0;

    } while(sqlite3_step(stmt) == SQLITE_ROW);

    data->rows = rows;

    sqlite3_finalize(stmt);

    return OPENCL_SUCCESS;
}

int
opencl_shard_data(opencl_sqlite_context *s, opencl_data_shard *shard, int column) {
	int rows = 0;
	int rowoffset=0;
	cl_int error=0;
    opencl_data *data = s->data_cpu;
    int *d = (int*)s->data_cpu->d;
	rows = data -> rows;
	shard->type = s->data_cpu->types[column];
	shard->rows = rows;
	int *v = malloc(rows * sizeof(int));
	int *cpu_map=v;
	//shard->d = clCreateBuffer(s->context, CL_MEM_ALLOC_HOST_PTR | CL_MEM_READ_ONLY, rows * sizeof(int), NULL, NULL);

	//int *destination = clEnqueueMapBuffer(s->queue, shard->d, CL_TRUE, CL_MAP_READ, 0, rows * sizeof(int), 0, NULL, NULL, &error);

    char *p =(char *) (d + column);

    do {
		//*destination = *((int *)p);
		*v = *((int *)p);

		/*
       	switch(data->types[i]) {
           	case OPENCL_INT :
              	break;
           	case OPENCL_INT64 :
               	break;
           	case OPENCL_FLOAT :
               	break;
          	case OPENCL_DOUBLE :
               	break;
           	default:
               	return OPENCL_ERR_TYPE;
		}*/
		// destination++;
		v++;
		p += data->stride;
		rows--;

    } while(rows);
	shard->d = clCreateBuffer(s->context,  CL_MEM_COPY_HOST_PTR | CL_MEM_READ_ONLY, s->data_cpu->rows * sizeof(int), cpu_map, NULL);

//	clEnqueueUnmapMemObject(s->queue, shard->d, destination, 0, NULL, NULL);

	return OPENCL_SUCCESS;
}